@kevinrabun/judges 3.34.1 → 3.36.0

package/dist/commands/benchmark-ai-output.js

@@ -0,0 +1,366 @@
+/**
+ * AI-output-specific benchmark cases targeting patterns characteristic of
+ * LLM-generated code: logic inversions, dead code, name-body mismatches,
+ * empty error handlers, and inadequate tests.
+ *
+ * Covers LOGIC, TEST, and enhanced HALLU prefixes.
+ */
+export const BENCHMARK_AI_OUTPUT = [
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  LOGIC — Inverted security conditions (AI frequently gets these wrong)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-logic-inverted-auth-check",
+        description: "AI inverts authentication check — grants access when NOT authenticated",
+        language: "typescript",
+        code: `export function authMiddleware(req: Request, res: Response, next: NextFunction) {
+  const token = req.headers.authorization;
+  const isAuthenticated = verifyToken(token);
+
+  if (!isAuthenticated) {
+    // AI inverted: should deny, but grants access
+    next();
+  } else {
+    res.status(401).json({ error: "Unauthorized" });
+  }
+}`,
+        expectedRuleIds: ["LOGIC-001", "AUTH-001"],
+        category: "ai-logic-error",
+        difficulty: "medium",
+        aiSource: "gpt-4",
+    },
+    {
+        id: "ai-logic-inverted-admin-check",
+        description: "AI inverts admin role check — grants admin when NOT admin",
+        language: "python",
+        code: `def require_admin(func):
+    def wrapper(request, *args, **kwargs):
+        if not request.user.is_admin:
+            return func(request, *args, **kwargs)  # Grants access when NOT admin
+        return HttpResponseForbidden("Admin access required")
+    return wrapper`,
+        expectedRuleIds: ["LOGIC-001", "AUTH-001"],
+        category: "ai-logic-error",
+        difficulty: "medium",
+        aiSource: "claude",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  LOGIC — Off-by-one errors (classic AI mistake)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-logic-off-by-one-loop",
+        description: "AI uses <= instead of < for array bounds",
+        language: "typescript",
+        code: `export function processItems(items: string[]) {
+  const results: string[] = [];
+  for (let i = 0; i <= items.length; i++) {
+    results.push(items[i].toUpperCase());
+  }
+  return results;
+}`,
+        expectedRuleIds: ["LOGIC-002"],
+        category: "ai-logic-error",
+        difficulty: "easy",
+        aiSource: "copilot",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  LOGIC — Dead code after return (AI generates unreachable code)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-logic-dead-code-after-return",
+        description: "AI generates cleanup code after a return statement",
+        language: "typescript",
+        code: `export function getUserProfile(userId: string) {
+  const user = db.findUser(userId);
+  if (!user) {
+    return null;
+    console.log("User not found, cleaning up...");
+    cache.invalidate(userId);
+  }
+  return user;
+}`,
+        expectedRuleIds: ["LOGIC-003"],
+        category: "ai-logic-error",
+        difficulty: "easy",
+        aiSource: "gpt-4",
+    },
+    {
+        id: "ai-logic-dead-code-after-throw",
+        description: "AI generates code after throw statement",
+        language: "typescript",
+        code: `export function validateInput(input: string) {
+  if (!input) {
+    throw new Error("Input is required");
+    logger.error("Validation failed for empty input");
+    metrics.increment("validation.failures");
+  }
+  return sanitize(input);
+}`,
+        expectedRuleIds: ["LOGIC-003"],
+        category: "ai-logic-error",
+        difficulty: "easy",
+        aiSource: "claude",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  LOGIC — Name-body mismatch (AI names don't match behavior)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-logic-name-mismatch-validate",
+        description: "Function named 'validate' but never rejects invalid input",
+        language: "typescript",
+        code: `export function validateUserInput(data: any) {
+  const name = data.name || "Anonymous";
+  const email = data.email || "unknown@example.com";
+  const age = data.age || 0;
+  return { name, email, age };
+}`,
+        expectedRuleIds: ["LOGIC-004"],
+        category: "ai-logic-error",
+        difficulty: "medium",
+        aiSource: "copilot",
+    },
+    {
+        id: "ai-logic-name-mismatch-delete",
+        description: "Function named 'delete' but only soft-deletes without actual removal",
+        language: "python",
+        code: `def delete_user(user_id):
+    user = User.query.get(user_id)
+    user.status = "inactive"
+    user.updated_at = datetime.now()
+    db.session.commit()
+    return {"message": "User updated"}`,
+        expectedRuleIds: ["LOGIC-004"],
+        category: "ai-logic-error",
+        difficulty: "medium",
+        aiSource: "gpt-4",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  LOGIC — Swapped comparison operands
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-logic-swapped-operands",
+        description: "AI swaps password and username in comparison",
+        language: "typescript",
+        code: `export async function login(username: string, password: string) {
+  const user = await db.findUser(username);
+  if (!user) return { error: "User not found" };
+
+  // AI swapped: comparing password to username
+  if (password === username) {
+    return { token: createToken(user) };
+  }
+  return { error: "Invalid credentials" };
+}`,
+        expectedRuleIds: ["LOGIC-005"],
+        category: "ai-logic-error",
+        difficulty: "hard",
+        aiSource: "claude",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  LOGIC — Empty catch blocks (AI silently swallows errors)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-logic-empty-catch",
+        description: "AI generates try/catch that silently swallows critical errors",
+        language: "typescript",
+        code: `export async function transferFunds(from: string, to: string, amount: number) {
+  try {
+    await db.beginTransaction();
+    await db.debit(from, amount);
+    await db.credit(to, amount);
+    await db.commit();
+  } catch (e) {
+  }
+  return { success: true };
+}`,
+        expectedRuleIds: ["LOGIC-006"],
+        category: "ai-logic-error",
+        difficulty: "easy",
+        aiSource: "copilot",
+    },
+    {
+        id: "ai-logic-empty-except-python",
+        description: "AI generates bare except that catches everything including SystemExit",
+        language: "python",
+        code: `def process_payment(order_id, amount):
+    try:
+        gateway = PaymentGateway()
+        result = gateway.charge(amount)
+        update_order(order_id, result)
+    except:
+        pass
+    return {"status": "processed"}`,
+        expectedRuleIds: ["LOGIC-006"],
+        category: "ai-logic-error",
+        difficulty: "easy",
+        aiSource: "gpt-4",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  TEST — Happy-path-only tests (AI generates tests without error cases)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-test-happy-path-only",
+        description: "AI generates multiple test cases but all are happy paths",
+        language: "typescript",
+        code: `describe("UserService", () => {
+  it("should create a user with valid data", () => {
+    const user = createUser({ name: "Alice", email: "alice@example.com" });
+    expect(user.id).toBeDefined();
+  });
+
+  it("should find a user by id", () => {
+    const user = findUser("123");
+    expect(user.name).toBe("Alice");
+  });
+
+  it("should update a user name", () => {
+    const user = updateUser("123", { name: "Bob" });
+    expect(user.name).toBe("Bob");
+  });
+
+  it("should list all users", () => {
+    const users = listUsers();
+    expect(users.length).toBeGreaterThan(0);
+  });
+});`,
+        expectedRuleIds: ["TEST-010"],
+        category: "ai-test-quality",
+        difficulty: "medium",
+        aiSource: "copilot",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  TEST — Status-code-only assertions (AI misses body validation)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-test-status-code-only",
+        description: "AI generates API tests that only check status codes",
+        language: "typescript",
+        code: `describe("API endpoints", () => {
+  it("GET /users returns 200", async () => {
+    const res = await request(app).get("/users");
+    expect(res.status).toBe(200);
+  });
+
+  it("POST /users returns 201", async () => {
+    const res = await request(app).post("/users").send({ name: "Test" });
+    expect(res.statusCode).toBe(201);
+  });
+
+  it("DELETE /users/1 returns 204", async () => {
+    const res = await request(app).delete("/users/1");
+    expect(res.status).toBe(204);
+  });
+});`,
+        expectedRuleIds: ["TEST-011"],
+        category: "ai-test-quality",
+        difficulty: "medium",
+        aiSource: "gpt-4",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  HALLU — Dependency confusion patterns
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-hallu-dependency-confusion",
+        description: "AI generates import of internal-looking unscoped package name",
+        language: "typescript",
+        code: `import { authenticate } from "auth-service-internal";
+import { getConfig } from "config-core-api";
+import { logger } from "logging-backend-sdk";
+
+export async function handleRequest(req: Request) {
+  const config = getConfig("production");
+  const user = await authenticate(req.headers.authorization);
+  logger.info("Request handled", { userId: user.id });
+  return { status: "ok" };
+}`,
+        expectedRuleIds: ["HALLU-001"],
+        category: "ai-dependency-confusion",
+        difficulty: "hard",
+        aiSource: "claude",
+    },
+    // ═══════════════════════════════════════════════════════════════════════════
+    //  NEGATIVE — Clean AI-generated code (should NOT flag)
+    // ═══════════════════════════════════════════════════════════════════════════
+    {
+        id: "ai-negative-clean-auth-middleware",
+        description: "Correct authentication middleware — should NOT flag logic errors",
+        language: "typescript",
+        code: `export function authMiddleware(req: Request, res: Response, next: NextFunction) {
+  const token = req.headers.authorization?.replace("Bearer ", "");
+  if (!token) {
+    res.status(401).json({ error: "No token provided" });
+    return;
+  }
+
+  try {
+    const decoded = verifyToken(token);
+    req.user = decoded;
+    next();
+  } catch (error) {
+    res.status(401).json({ error: "Invalid token" });
+  }
+}`,
+        expectedRuleIds: [],
+        unexpectedRuleIds: ["LOGIC-001", "LOGIC-005", "LOGIC-006"],
+        category: "ai-negative",
+        difficulty: "easy",
+        aiSource: "gpt-4",
+    },
+    {
+        id: "ai-negative-clean-error-handling",
+        description: "Proper error handling with logging — should NOT flag empty catch",
+        language: "typescript",
+        code: `export async function processOrder(orderId: string) {
+  try {
+    const order = await db.findOrder(orderId);
+    if (!order) throw new Error("Order not found");
+
+    await paymentGateway.charge(order.amount);
+    await db.updateOrder(orderId, { status: "paid" });
+
+    return { success: true };
+  } catch (error) {
+    logger.error("Order processing failed", { orderId, error });
+    await db.updateOrder(orderId, { status: "failed" });
+    throw error;
+  }
+}`,
+        expectedRuleIds: [],
+        unexpectedRuleIds: ["LOGIC-003", "LOGIC-006"],
+        category: "ai-negative",
+        difficulty: "easy",
+        aiSource: "claude",
+    },
+    {
+        id: "ai-negative-clean-validation",
+        description: "Proper input validation function — should NOT flag name mismatch",
+        language: "typescript",
+        code: `export function validateUserInput(data: unknown): { valid: boolean; errors: string[] } {
+  const errors: string[] = [];
+
+  if (!data || typeof data !== "object") {
+    return { valid: false, errors: ["Input must be an object"] };
+  }
+
+  const { name, email, age } = data as Record<string, unknown>;
+
+  if (typeof name !== "string" || name.length < 2) {
+    errors.push("Name must be at least 2 characters");
+  }
+  if (typeof email !== "string" || !email.includes("@")) {
+    errors.push("Invalid email format");
+  }
+  if (typeof age !== "number" || age < 0 || age > 150) {
+    errors.push("Age must be between 0 and 150");
+  }
+
+  return { valid: errors.length === 0, errors };
+}`,
+        expectedRuleIds: [],
+        unexpectedRuleIds: ["LOGIC-004"],
+        category: "ai-negative",
+        difficulty: "easy",
+        aiSource: "copilot",
+    },
+];
+//# sourceMappingURL=benchmark-ai-output.js.map

package/dist/commands/benchmark-ai-output.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"benchmark-ai-output.js","sourceRoot":"","sources":["../../src/commands/benchmark-ai-output.ts"],"names":[],"mappings":"AAEA;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,mBAAmB,GAAoB;IAClD,8EAA8E;IAC9E,yEAAyE;IACzE,8EAA8E;IAC9E;QACE,EAAE,EAAE,8BAA8B;QAClC,WAAW,EAAE,wEAAwE;QACrF,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;;EAUR;QACE,eAAe,EAAE,CAAC,WAAW,EAAE,UAAU,CAAC;QAC1C,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,QAAQ;QACpB,QAAQ,EAAE,OAAO;KAClB;IACD;QACE,EAAE,EAAE,+BAA+B;QACnC,WAAW,EAAE,2DAA2D;QACxE,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE;;;;;mBAKS;QACf,eAAe,EAAE,CAAC,WAAW,EAAE,UAAU,CAAC;QAC1C,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,QAAQ;QACpB,QAAQ,EAAE,QAAQ;KACnB;IAED,8EAA8E;IAC9E,kDAAkD;IAClD,8EAA8E;IAC9E;QACE,EAAE,EAAE,0BAA0B;QAC9B,WAAW,EAAE,0CAA0C;QACvD,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;EAMR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,SAAS;KACpB;IAED,8EAA8E;IAC9E,kEAAkE;IAClE,8EAA8E;IAC9E;QACE,EAAE,EAAE,iCAAiC;QACrC,WAAW,EAAE,oDAAoD;QACjE,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;EAQR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,OAAO;KAClB;IACD;QACE,EAAE,EAAE,gCAAgC;QACpC,WAAW,EAAE,yCAAyC;QACtD,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;EAOR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;KACnB;IAED,8EAA8E;IAC9E,8DAA8D;IAC9D,8EAA8E;IAC9E;QACE,EAAE,EAAE,iCAAiC;QACrC,WAAW,EAAE,2DAA2D;QACxE,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;EAKR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,QAAQ;QACpB,QAAQ,EAAE,SAAS;KACpB;IACD;QACE,EAAE,EAAE,+BAA+B;QACnC,WAAW,EAAE,sEAAsE;QACnF,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE;;;;;uCAK6B;QACnC,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,QAAQ;QACpB,QAAQ,EAAE,OAAO;KAClB;IAED,8EAA8E;IAC9E,uCAAuC;IACvC,8EAA8E;IAC9E;QACE,EAAE,EAAE,2BAA2B;QAC/B,WAAW,EAAE,8CAA8C;QAC3D,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;EASR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;KACnB;IAED,8EAA8E;IAC9E,4DAA4D;IAC5D,8EAA8E;IAC9E;QACE,EAAE,EAAE,sBAAsB;QAC1B,WAAW,EAAE,+DAA+D;QAC5E,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;EASR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,SAAS;KACpB;IACD;QACE,EAAE,EAAE,8BAA8B;QAClC,WAAW,EAAE,uEAAuE;QACpF,QAAQ,EAAE,QAAQ;QAClB,IAAI,EAAE;;;;;;;mCAOyB;QAC/B,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,gBAAgB;QAC1B,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,OAAO;KAClB;IAED,8EAA8E;IAC9E,yEAAyE;IACzE,8EAA8E;IAC9E;QACE,EAAE,EAAE,yBAAyB;QAC7B,WAAW,EAAE,0DAA0D;QACvE,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;;;;;;;;;;;;IAoBN;QACA,eAAe,EAAE,CAAC,UAAU,CAAC;QAC7B,QAAQ,EAAE,iBAAiB;QAC3B,UAAU,EAAE,QAAQ;QACpB,QAAQ,EAAE,SAAS;KACpB;IAED,8EAA8E;IAC9E,kEAAkE;IAClE,8EAA8E;IAC9E;QACE,EAAE,EAAE,0BAA0B;QAC9B,WAAW,EAAE,qDAAqD;QAClE,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;;;;;;;IAeN;QACA,eAAe,EAAE,CAAC,UAAU,CAAC;QAC7B,QAAQ,EAAE,iBAAiB;QAC3B,UAAU,EAAE,QAAQ;QACpB,QAAQ,EAAE,OAAO;KAClB;IAED,8EAA8E;IAC9E,yCAAyC;IACzC,8EAA8E;IAC9E;QACE,EAAE,EAAE,+BAA+B;QACnC,WAAW,EAAE,+DAA+D;QAC5E,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;EASR;QACE,eAAe,EAAE,CAAC,WAAW,CAAC;QAC9B,QAAQ,EAAE,yBAAyB;QACnC,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;KACnB;IAED,8EAA8E;IAC9E,wDAAwD;IACxD,8EAA8E;IAC9E;QACE,EAAE,EAAE,mCAAmC;QACvC,WAAW,EAAE,kEAAkE;QAC/E,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;;;;;;EAcR;QACE,eAAe,EAAE,EAAE;QACnB,iBAAiB,EAAE,CAAC,WAAW,EAAE,WAAW,EAAE,WAAW,CAAC;QAC1D,QAAQ,EAAE,aAAa;QACvB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,OAAO;KAClB;IACD;QACE,EAAE,EAAE,kCAAkC;QACtC,WAAW,EAAE,kEAAkE;QAC/E,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;;;;;;EAcR;QACE,eAAe,EAAE,EAAE;QACnB,iBAAiB,EAAE,CAAC,WAAW,EAAE,WAAW,CAAC;QAC7C,QAAQ,EAAE,aAAa;QACvB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,QAAQ;KACnB;IACD;QACE,EAAE,EAAE,8BAA8B;QAClC,WAAW,EAAE,kEAAkE;QAC/E,QAAQ,EAAE,YAAY;QACtB,IAAI,EAAE;;;;;;;;;;;;;;;;;;;;EAoBR;QACE,eAAe,EAAE,EAAE;QACnB,iBAAiB,EAAE,CAAC,WAAW,CAAC;QAChC,QAAQ,EAAE,aAAa;QACvB,UAAU,EAAE,MAAM;QAClB,QAAQ,EAAE,SAAS;KACpB;CACF,CAAC"}

package/dist/commands/benchmark.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"benchmark.d.ts","sourceRoot":"","sources":["../../src/commands/benchmark.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAMH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAa3C,MAAM,WAAW,aAAa;IAC5B,2CAA2C;IAC3C,EAAE,EAAE,MAAM,CAAC;IACX,mDAAmD;IACnD,WAAW,EAAE,MAAM,CAAC;IACpB,2BAA2B;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,IAAI,EAAE,MAAM,CAAC;IACb,2DAA2D;IAC3D,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,mEAAmE;IACnE,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC7B,kEAAkE;IAClE,QAAQ,EAAE,MAAM,CAAC;IACjB,uBAAuB;IACvB,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,MAAM,CAAC;IACvC,iFAAiF;IACjF,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,uBAAuB;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,6BAA6B;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,2BAA2B;IAC3B,UAAU,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,QAAQ,EAAE,MAAM,CAAC;IACjB,gDAAgD;IAChD,MAAM,EAAE,MAAM,CAAC;IACf,+CAA+C;IAC/C,aAAa,EAAE,MAAM,CAAC;IACtB,2DAA2D;IAC3D,aAAa,EAAE,MAAM,CAAC;IACtB,0DAA0D;IAC1D,cAAc,EAAE,MAAM,CAAC;IACvB,8DAA8D;IAC9D,cAAc,EAAE,MAAM,CAAC;IACvB,wDAAwD;IACxD,SAAS,EAAE,MAAM,CAAC;IAClB,qDAAqD;IACrD,MAAM,EAAE,MAAM,CAAC;IACf,qEAAqE;IACrE,OAAO,EAAE,MAAM,CAAC;IAChB,mDAAmD;IACnD,aAAa,EAAE,MAAM,CAAC;IACtB,iDAAiD;IACjD,mBAAmB,EAAE,MAAM,CAAC;IAC5B,wDAAwD;IACxD,oBAAoB,EAAE,MAAM,CAAC;IAC7B,oEAAoE;IACpE,eAAe,EAAE,MAAM,CAAC;IACxB,iEAAiE;IACjE,YAAY,EAAE,MAAM,CAAC;IACrB,8CAA8C;IAC9C,aAAa,EAAE,MAAM,CAAC;IACtB,2BAA2B;IAC3B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IAC5C,wBAAwB;IACxB,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,oBAAoB,CAAC,CAAC;IAC/C,+BAA+B;IAC/B,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;IAChD,8DAA8D;IAC9D,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IAC7C,8BAA8B;IAC9B,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,MAAM,WAAW,gBAAgB;IAC/B,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,OAAO,CAAC;IAChB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,oBAAoB,EAAE,MAAM,EAAE,CAAC;IAC/B,QAAQ,CAAC,EAAE,OAAO,EAAE,CAAC;CACtB;AAID,eAAO,MAAM,eAAe,EAAE,aAAa,EA4qE1C,CAAC;AAIF,wBAAgB,iBAAiB,CAAC,KAAK,CAAC,EAAE,aAAa,EAAE,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,eAAe,CA0P5F;AAID,MAAM,WAAW,oBAAoB;IACnC,2CAA2C;IAC3C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,4CAA4C;IAC5C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,yCAAyC;IACzC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iDAAiD;IACjD,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,eAAe,CAAC;CAC5B;AAED,MAAM,WAAW,mBAAmB;IAClC,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,eAAe,CAAC;CACzB;AAED;;;;;;;;;GASG;AACH,wBAAgB,aAAa,CAAC,OAAO,GAAE,oBAAyB,GAAG,mBAAmB,CA6CrF;AAID,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,eAAe,GAAG,MAAM,CAsFrE;AAID,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,eAAe,GAAG,MAAM,CA8IvE;AAID,wBAAgB,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,IAAI,CAiRjD;AAOD,MAAM,WAAW,kBAAkB;IACjC,mDAAmD;IACnD,mBAAmB,EAAE,MAAM,CAAC;IAC5B,gFAAgF;IAChF,WAAW,EAAE,MAAM,CAAC;IACpB,2DAA2D;IAC3D,cAAc,EAAE,MAAM,CAAC;IACvB,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC1C,6BAA6B;IAC7B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAAC;IAChD,+BAA+B;IAC/B,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACnG,0DAA0D;IAC1D,kBAAkB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CAC9C;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,yDAAyD;IACzD,cAAc,EAAE,MAAM,CAAC;IACvB,gDAAgD;IAChD,WAAW,EAAE,OAAO,CAAC;IACrB,gDAAgD;IAChD,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;CACtB;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,eAAe,GAAG,kBAAkB,CA0E7E;AAED;;GAEG;AACH,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,kBAAkB,GAAG,MAAM,CAuF3E;AAOD,MAAM,WAAW,mBAAmB;IAClC,qCAAqC;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,2BAA2B;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,0CAA0C;IAC1C,QAAQ,EAAE,KAAK,CAAC;QAAE,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CACrC;AAED;;;;;;GAMG;AACH,wBAAgB,8BAA8B,CAAC,MAAM,EAAE,mBAAmB,EAAE,GAAG,aAAa,EAAE,CA6B7F;AAED;;;;;GAKG;AACH,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,aAAa,EAAE,EAAE,UAAU,EAAE,aAAa,EAAE,GAAG,aAAa,EAAE,CAM9G"}
+{"version":3,"file":"benchmark.d.ts","sourceRoot":"","sources":["../../src/commands/benchmark.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAMH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,aAAa,CAAC;AAc3C,MAAM,WAAW,aAAa;IAC5B,2CAA2C;IAC3C,EAAE,EAAE,MAAM,CAAC;IACX,mDAAmD;IACnD,WAAW,EAAE,MAAM,CAAC;IACpB,2BAA2B;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,IAAI,EAAE,MAAM,CAAC;IACb,2DAA2D;IAC3D,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,mEAAmE;IACnE,iBAAiB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC7B,kEAAkE;IAClE,QAAQ,EAAE,MAAM,CAAC;IACjB,uBAAuB;IACvB,UAAU,EAAE,MAAM,GAAG,QAAQ,GAAG,MAAM,CAAC;IACvC,iFAAiF;IACjF,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,uBAAuB;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,6BAA6B;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,2BAA2B;IAC3B,UAAU,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,QAAQ,EAAE,MAAM,CAAC;IACjB,gDAAgD;IAChD,MAAM,EAAE,MAAM,CAAC;IACf,+CAA+C;IAC/C,aAAa,EAAE,MAAM,CAAC;IACtB,2DAA2D;IAC3D,aAAa,EAAE,MAAM,CAAC;IACtB,0DAA0D;IAC1D,cAAc,EAAE,MAAM,CAAC;IACvB,8DAA8D;IAC9D,cAAc,EAAE,MAAM,CAAC;IACvB,wDAAwD;IACxD,SAAS,EAAE,MAAM,CAAC;IAClB,qDAAqD;IACrD,MAAM,EAAE,MAAM,CAAC;IACf,qEAAqE;IACrE,OAAO,EAAE,MAAM,CAAC;IAChB,mDAAmD;IACnD,aAAa,EAAE,MAAM,CAAC;IACtB,iDAAiD;IACjD,mBAAmB,EAAE,MAAM,CAAC;IAC5B,wDAAwD;IACxD,oBAAoB,EAAE,MAAM,CAAC;IAC7B,oEAAoE;IACpE,eAAe,EAAE,MAAM,CAAC;IACxB,iEAAiE;IACjE,YAAY,EAAE,MAAM,CAAC;IACrB,8CAA8C;IAC9C,aAAa,EAAE,MAAM,CAAC;IACtB,2BAA2B;IAC3B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IAC5C,wBAAwB;IACxB,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,oBAAoB,CAAC,CAAC;IAC/C,+BAA+B;IAC/B,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,CAAC,CAAC;IAChD,8DAA8D;IAC9D,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;IAC7C,8BAA8B;IAC9B,KAAK,EAAE,UAAU,EAAE,CAAC;CACrB;AAED,MAAM,WAAW,gBAAgB;IAC/B,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,CAAC;IACtB,cAAc,EAAE,MAAM,CAAC;IACvB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,MAAM,CAAC;IACf,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,UAAU;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,OAAO,CAAC;IAChB,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,oBAAoB,EAAE,MAAM,EAAE,CAAC;IAC/B,QAAQ,CAAC,EAAE,OAAO,EAAE,CAAC;CACtB;AAID,eAAO,MAAM,eAAe,EAAE,aAAa,EA6qE1C,CAAC;AAIF,wBAAgB,iBAAiB,CAAC,KAAK,CAAC,EAAE,aAAa,EAAE,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,eAAe,CA0P5F;AAID,MAAM,WAAW,oBAAoB;IACnC,2CAA2C;IAC3C,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,4CAA4C;IAC5C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,yCAAyC;IACzC,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,iDAAiD;IACjD,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,8DAA8D;IAC9D,QAAQ,CAAC,EAAE,eAAe,CAAC;CAC5B;AAED,MAAM,WAAW,mBAAmB;IAClC,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,MAAM,EAAE,eAAe,CAAC;CACzB;AAED;;;;;;;;;GASG;AACH,wBAAgB,aAAa,CAAC,OAAO,GAAE,oBAAyB,GAAG,mBAAmB,CA6CrF;AAID,wBAAgB,qBAAqB,CAAC,MAAM,EAAE,eAAe,GAAG,MAAM,CAsFrE;AAID,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,eAAe,GAAG,MAAM,CA8IvE;AAID,wBAAgB,YAAY,CAAC,IAAI,EAAE,MAAM,EAAE,GAAG,IAAI,CAiRjD;AAOD,MAAM,WAAW,kBAAkB;IACjC,mDAAmD;IACnD,mBAAmB,EAAE,MAAM,CAAC;IAC5B,gFAAgF;IAChF,WAAW,EAAE,MAAM,CAAC;IACpB,2DAA2D;IAC3D,cAAc,EAAE,MAAM,CAAC;IACvB,6DAA6D;IAC7D,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,eAAe,CAAC,CAAC;IAC1C,6BAA6B;IAC7B,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAAC,CAAC;IAChD,+BAA+B;IAC/B,aAAa,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,cAAc,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACnG,0DAA0D;IAC1D,kBAAkB,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;CAC9C;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,yDAAyD;IACzD,cAAc,EAAE,MAAM,CAAC;IACvB,gDAAgD;IAChD,WAAW,EAAE,OAAO,CAAC;IACrB,gDAAgD;IAChD,YAAY,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,EAAE,MAAM,CAAC;IACjB,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,YAAY,EAAE,MAAM,CAAC;CACtB;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,eAAe,GAAG,kBAAkB,CA0E7E;AAED;;GAEG;AACH,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,kBAAkB,GAAG,MAAM,CAuF3E;AAOD,MAAM,WAAW,mBAAmB;IAClC,qCAAqC;IACrC,IAAI,EAAE,MAAM,CAAC;IACb,2BAA2B;IAC3B,QAAQ,EAAE,MAAM,CAAC;IACjB,0CAA0C;IAC1C,QAAQ,EAAE,KAAK,CAAC;QAAE,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;CACrC;AAED;;;;;;GAMG;AACH,wBAAgB,8BAA8B,CAAC,MAAM,EAAE,mBAAmB,EAAE,GAAG,aAAa,EAAE,CA6B7F;AAED;;;;;GAKG;AACH,wBAAgB,sBAAsB,CAAC,QAAQ,EAAE,aAAa,EAAE,EAAE,UAAU,EAAE,aAAa,EAAE,GAAG,aAAa,EAAE,CAM9G"}

package/dist/commands/benchmark.js

@@ -23,6 +23,7 @@ import { BENCHMARK_INFRASTRUCTURE } from "./benchmark-infrastructure.js";
 import { BENCHMARK_COMPLIANCE_ETHICS } from "./benchmark-compliance-ethics.js";
 import { BENCHMARK_AI_AGENTS } from "./benchmark-ai-agents.js";
 import { BENCHMARK_ADVANCED_CASES } from "./benchmark-advanced.js";
+import { BENCHMARK_AI_OUTPUT } from "./benchmark-ai-output.js";
 // ─── Built-in Benchmark Cases ───────────────────────────────────────────────
 export const BENCHMARK_CASES = [
     // ── SQL Injection ──
@@ -2171,6 +2172,7 @@ export function UserList({ users, onSelect, searchLabel = "Search users" }: User
     ...BENCHMARK_COMPLIANCE_ETHICS,
     ...BENCHMARK_AI_AGENTS,
     ...BENCHMARK_ADVANCED_CASES,
+    ...BENCHMARK_AI_OUTPUT,
 ];
 // ─── Benchmark Runner ───────────────────────────────────────────────────────
 export function runBenchmarkSuite(cases, judgeId) {